Towards Privacy Preserving and Efficiency in Fog Selection for Federated Learning

Abstract

Federated learning (FL) is an emerging trend related to the concept of distributed Machine Learning (ML). It focuses on a collaborative training process locally conducted on the dataset of the client devices in order to preserve the users’ privacy. Nonetheless, this solution still suffers from many challenges dealing with privacy, security, and performance. In this work, we introduce a novel policy-based FL approach for improving privacy, security, and performance in federated learning. Our proposed solution ensures reliability, communications security, and heterogeneous privacy (i.e., the users have different privacy attitudes and expectations). In addition, it guarantees performance in terms of the dataset’s quality and scalability. To prove the effectiveness of our model, we perform a security and performance evaluation by assuming a threat model with attackers having different behaviors.